My parallel stereo single ended triode is now ready :)

[G.Castle]

Not set up your models yet Gabe, but did try your modification in isolation on the first two stages in simulation with what was available to me.

The result would be in agreement: the only thing that the real world could throw up from it, is the second stage of the voltage amplifier going unstable: but it's unlikely to take off in reality particularly with those grid stoppers. It seems odd that the first stage is the culprit.

[kalee20]

Originally Posted by Gabe001

So, here is an idea to address the HF loss.

I'm away right now, just about keeping up! And watching the thread with interest.

But, although it's laudable to improve HF response, I'm more interested why it's not in accordance with theory... when that's understood, the fix may become apparent, rather than adding a HF boost somewhere else. Unfortunately I'm not going to be able to run any LTSpice simulation till Tuesday at least!

[trobbins]

Gabe, your REW plot in post #2 indicated you were only using a 48kHz sampling rate. Afaik (https://mnaganov.github.io/2018/05/m...er-uca202.html) that UCA-202 has a 96kHz sampling rate for up to 48kHz bandwidth. Have you been able to use the higher sampling rate, with an ASIO driver, and do a loopback test and then apply a calibration to remove any measurement system bandwidth roll-offs from impacting on your measurements of the amp?

I find such a soundcard/REW bench tool very helpful when making part value changes to an amp to diagnose how bandwidth is impacted, as a frequency response sweep is just so easy to make. Although 48kHz bandwidth is still rather limiting, and even a 96kHz bandwidth can be restrictive with my soundcard, it does allow valuable information on gain and phase shifts to be observed within the applicable bandwidth and hence to relate them back to circuit assessment.

PS. you can't use the squarewave output from such a soundcard system as a valid test signal for stability assessment (or even bandwidth for that matter), due to the bandwidth limit of the soundcard's generated signal.

PSS. Sorry I see the UCA-202 only has up to 48kHz sampling, so 24kHz bandwidth max, but still worthwile performing a loopback and calibration to see how 'flat' you can make the measurement system.

[Jez1234]

I still use and trust an all analogue measurement set up in all my work... and IMHO for good reasons. It's the real thing in real time and not some interpretation and extrapolation of what the software believes you should be seeing.. to put it as a somewhat crude no doubt over-simplification. It's not just "rubbish in = rubbish out" but there is a second and higher orders effect of what the computer makes of the rubbish and what it decides is the right way to report this to you... If you see what I mean! You really do need to apply a sanity check to it to be sure.

[trobbins]

Same for simulation using eg. LTSpice, and also for analog when measurement equipment is not functioning correctly or the user hasn't appreciated the limits of the measurement system. It is up to the user to double check their measurement equipment and setup to have valid confidence in reporting results, or have forum members make note of such issues

[Jez1234]

Originally Posted by trobbins

Same for simulation using eg. LTSpice, and also for analog when measurement equipment is not functioning correctly or the user hasn't appreciated the limits of the measurement system. It is up to the user to double check their measurement equipment and setup to have valid confidence in reporting results, or have forum members make note of such issues

"also for analog when measurement equipment is not functioning correctly" well let us assume it is functioning correctly at least!

You know what I mean I'm sure... ie that with digital you are not seeing a real time "first principles" result but a result derived from the input being digitised and those numbers being fed through algorithms to give an ultimate interpolation of what it is believed those numbers probably mean based on what was sampled at a specific instance in time and on whether Hanning, Hamming, Blackman, Tukey, Cooley or Tom Dick and Harry had the right idea....

I'll argue till the cows come home that "the old ways were best" and the acolytes of the byte will argue the opposite til the cows come home, go away again for a while, and eventually become bored with the whole thing and wish they were horses... so to avoid further thread drift, I have nothing more to say

[Radio Wrangler]

When you are doing reasonably demanding measurements (analogue or digital) on systems involving lowpass responses (intentional or otherwise) you find that bandwidth limits are inevitable. Unlimited bandwidth is impossible. There will always be limitations imposed on Fourier series by filters, roll-off and limited scanning boundaries.

The truncation of spectra or the truncation od Fourier series gives rise to "Gibbs' Oscillations" Which may not be oscillations lke those produced by an oscillator, but ar rather ripple effects seen in responses when the Fourier series of a known easy-to-think-of waveform like a square-wave is cut off. The truncated Fourier series sums up to create what looks like the intended waveform, but with ringing. A right royal pain if you'r looking for ringing! It also messes with flatness plots. The 'Oscillations' bit is factually correct, but often misleading. I think Wikipedia titles their article "Gibbs' Phenomena'

If you use soundcards as your signal source, you gett a Gibbs' from the anti-alias filtering.... The analogue final anti-alias is obvious, you can see it on the soundcard schematic, but don't forget the digital filtering preceding the DAC for oversampling. This also truncates the ideal Fourier series (spectrum) and gives its own Gibbs' phenomenon.

This is more often a peril with performing digital-assisted analogue parameter measurement, but effects are still there with pure-analogue test gear.

Of course there are also roll-offs and sample rates implicit in performing simulations and you get Gibbs' ripples there too. With modern simulators you often get fancy automatic control of increment rates and the Gibbs' phenomena go all over the shop, getting changes throughout the simulation run.

In the analogue world, as your fundamentals go up in frequency more and more of the harmonic structure falls off the top of the lowpass nature of pretty much everything and you get Gibbs' phenomena.

If a Gibbs' ripple is caused by the bandwidth liitations of the DUT, then what you see is real and legitimate. This DUT will make those ripples in real world use. But those Gibbs' ripples resulting from bandwidth limitation in the test gear are artefacts of the test gear and won't be present when you screw the DUT into some final system for real use.

When you're doing scanned measurements, whether it's a real analogue scan, or boundaries implicit in the maths of a digital system you also get Gibbs' going on, but it's harder to get your head around.

There are other nasties to trip up the unwary in measuring things with high finesse. Noise-noise convolution effects on apparent noise floors is a right stinker.

So neither purely analogue testing nor DSP-based testing get everything right and impose no artefacts of their own. Each offers insights less obvious in the other world. The art is to do a bit of both and to know how to combine their strengths, not their weaknesses. Neither world has exclusive keys to truth.

The effect of simple head gap width on the frequency response in tape replay is a nice example of truncation in time giving a severe Gibbs' phenomenon in an entirely analogue world. In this case you get a full-blooded sinc function which you see if you look high enough in frequency.

David

[Gabe001]

A good example of Gibbs phenomenon when trying to generate a square wave digitally (via the DAC), from real life. I have a proper sine and square wave generator now (and also the Arduino one which works well) which do not exhibit this artefact. Took me a while to work this out, so hope someone finds this useful in future.

The FR lack of HF extension is "Mia culpa'. I should have known as I came across this before. I'll test it with scope and plot it on Andy's paper once I do the adjustments

I think the REW THD measurements are accurate. If there is anyone close by with a spectrum analyser I'd be happy to do a comparative check.

Kalee - thanks. I'll hold off doing any modification for now.

[G.Castle]

Still watching!

I'll throw my hat into the ring with real measurements in the real world, being the best way forward once there is hardware in existence.

However, if anyone has ever had a piece of circuitry differentiate the manufactured sine wave that some function generators throw out, they will know what real frustration is.

Having suitable trustworthy test equipment and knowing its limitations is paramount.

Originally Posted by Gabe001

I think the REW THD measurements are accurate. If there is anyone close by with a spectrum analyser I'd be happy to do a comparative check.

I have so far failed with the sim, (will keep trying, it needs a sit down session with the computer, some of the lib files are duplicated and are causing conflict with the spice analysis), I am most interested in the correlation between different valve models and a real world thd+noise measurement.

Whilst a wide FR goes hand in hand with faithful reproduction of the signal as mentioned before, the chasing of amplifier superlatives particularly THD, while educational, seems a pointless errand beyond what can be detected by the human ear: when there is a loud speaker in between the amplifier and listener.

It could be claimed to add to the mellifluous character of the sound though, which is why I thought the topology was chosen in the first place and executed so very well in hardware.

[John10b]

Well done on completing your project, what an interesting read this has been, than you.
John

[kalee20]

Originally Posted by G.Castle

Still watching!

I'll throw my hat into the ring with real measurements in the real world, being the best way forward once there is hardware in existence.

Seconded!

LTSpice (other simulators are available) are powerful tools, but they're tools at best, and only models... and only give results according to data put in.

It would be difficult to model stray field from an output transformer interacting with electron flow in a badly-placed first-stage amplifier pentode, for instance...

They're best used to give confidence to build hardware... and then if there's discrepancies between hardware measurements and simulated results, much fun can be had reconciling them and maybe improving models. Conversely, if hardware misbehaves and it's hard to understand why, model it, add parasitic inductances, transit times... till the anomaly happens - it may be quicker than benchwork.

One big advantage of LTSpice is that the 'scope probe doesn't load the simulated circuit in any way at all!

[Gabe001]

I'll put the Lt spice and real life figures side by side once I improve the HF response. So far it's been incredibly accurate.

[G.Castle]

We are all waiting with our breath bated Gabe!

With reference to the Ayumi files in post 63, in some versions of Spice, I'm told they cannot be opened fom the .inc command.

What you can do in that event is to extract the zip containing file to a file, say on your desktop, the individual macromodel files have the extension .inc, however they can be individually opened with the spice IV program, then copied and paste into the relevant folder that can be found in:

C:\Programfiles\LTC\LTspice IV\lib\sub\*

*In my case it is the 'valves' file. This is also opened with the LTspice program and the copied macromodel is pasted in.

You must also remember that for each macro, in the sub file, there must also be an .asy file in the lib\sym\*

As there are unasigned symbols for triode, tetrode, and pentodes, in the lib, you can just open one of those and rename it the same as the macro you require is named, by right clicking on the graphic over the generic name and changing it in the dialog box. then just select file, then 'save as', again using the name exactly as appears in the macro.

After following the above, you should now have your selected device available whenever you go to the component select tab on the program, saving ever having to do a .inc command for that device.

There is no need to change anything within the macro text files, they are ready for use.

Just remember, most of the macros for pentodes are for use with the tetrode symbol as the suppressor is assumed in the Model to be connected to the cathode.

I've found the Ayumi macros to be the better of those available and worth the trouble to get them onto the LT spice program: there is also a greater choice to select from, if you do want to try some exotica.

Hope that helps.

[Gabe001]

Thanks for that. I'll give them a try. Should have something ready over the weekend.just waiting to see if kalee can propose a better HF solution to mine before I modify.

Also just figured out yesterday how to use Lt spice to work out output impedance and damping factor. DF works out at 5 for this circuit at 1kz, if I've done it correctly.

[G.Castle]

Originally Posted by Gabe001

Thanks for that. I'll give them a try. Should have something ready over the weekend.just waiting to see if kalee can propose a better HF solution to mine before I modify.

Also just figured out yesterday how to use Lt spice to work out output impedance and damping factor. DF works out at 5 for this circuit at 1kz, if I've done it correctly.

Good plan! Peter recently helped a friend of ours to fettle a little PP self split class A amplifier concept, the improvement was very creditable indeed, yours seem to be starting from the POV of being already very good though.

Your DF measurements seem a worthwhile option, I thought that 5 was very good. I keep forgetting that it has local NFB around the OP stage, so much chatter about other NFB free topologies up thread.

My previous post: I hope people will forgive the use of the term 'file' when what I really mean is 'folder'. Tiring day!

[Gabe001]

Yes, the local cathode feedback improves DF a bit. Typical triode DF is about 3 afaik which is why they work well enough without feedback. Of course you wouldn't want a THD of >5% as that's no longer music - just an intermodulation porridge.

I am curious how feedback less 300b triodes perform from the thd perspective at 7w output. I think they distort very little.

[G.Castle]

Off the top of my head, and from limited experience of the 300b as only ever a listener to the equipment of others all I can speculate is that the output is run with speaker mismatch and as the valve will have a quite low ra then it will be technically pretty good. I'd say the sound was fairly clean from the amp I'd listened to, but spec., of 5W from a 40W anode dissipation I'd expect it to be silky smooth if the design was up to scratch.

At £200 a pair for modern production examples, I'll be the Philistine and buy pairs of 6BW6 and use PP UL transformers with lashings of NFB!

(My daily listening is done with a stereo version of the Mullard 3/3 for context)

[Jez1234]

I've only ever heard one SET zero feedback amp that I consider listenable.... and it wasn't the Audio Note Ongaku which I thought was awful. A snip at £70,000! I think someone said something like "sounds mellow like grans old radio" or similar up thread and yes indeed that's the problem.

[G.Castle]

Originally Posted by Jez1234

I've only ever heard one SET zero feedback amp that I consider listenable.... and it wasn't the Audio Note Ongaku which I thought was awful.

I was shown a boutique amplifier once, (not SET), that was in for repair with Russell, a pro audio tech friend of mine.

The amp was touted as being a 'No NFB design' by the maker, Russell pointed out that the output transformers speaker windings were earthed to chassis via the bottom tags of the volume pots, that's strange in itself, but the tags were 'earthed' to the cover of the pots by connections via sleeving, then via braid to the chassis.

The thing is the other side of the transformer speaker winding was taken to chassis at the speaker connectors, (binding posts, the ofending ones screwed directly to the metalwork).

How could it work? The transformer would be shorted out!

The two little pieces of sleeving on the pots weren't covering wire, they were resistors!

No NFB my arse! OK, only a fixed amount and least effective at full wick but NFB nonetheless. Talk about not having the courage of your convictions.

Shorting the resistors out didn’t improve the sound.

[GrimJosef]

Originally Posted by Gabe001

... I am curious how feedback less 300b triodes perform from the thd perspective at 7w output. I think they distort very little.

The datasheet https://frank.pocnet.net/sheets/084/3/300B.pdf says on p2 that a single one at 7W has a THD of 5%, but on p5 there's a graph which covers what appears to be the same operating point (4k load, 350V on the anode, -74V grid bias) and it indicates 7W out and 2.5% THD. Still, even 2.5% should be clearly audible with decent speakers. So high fidelity (in the sense of output signal vs input signal) it ain't.

EDIT: See correction a few posts below.

Cheers,

GJ